Burke was responding to AI labs' complaints that Chinese companies are copying their models by training on their outputs. He asked listeners to take the geopolitics out of it and consider only the argument about intellectual property.
You know, not only that, but I also, I've had a, I've seen a lot of the Anthropic and these other LLM company, model company posts where they complain about how they, well, at least according to them, their models are being stolen. And they'll always point to China and the companies in China as stealing their models. And which is just like what people have always said forever. about manufacturing coming out of China in terms of sort of the, it's, it's like the scapegoat of, that's just what people point to is copycat. But the truth of the matter is these days, there's a lot of really good engineering coming out of China.
Yeah.
But what's interesting to me is how these LLM companies, they, they still draw upon this old image, uh, of China and its sourcing for what it it's IP, I guess. And these are companies like Anthropic and, you know, OpenAI and Meta who are in the business of stealing everything made by anybody ever. And they then have the gall to point at literally any other company and say, you're reverse engineering or stealing our IP. I just, I have a really hard time reconciling why I should feel bad. Like take the China and the US and the whatever out of it, right? Take all the geopolitics out. Why should I feel bad? If a company that has ingested all these authors and musicians and everyone else's work then feels like their IP for their LLM has been infringed upon by a competitor. You know, it's like, make me care.
Well, also, the other thing is, I think there's also a fairly large layer of xenophobia to it. I also think, paraphrasing my girlfriend here. America misses the USSR so bad. Oh, we miss having an argument with Russia. Oh, we don't have an adversary anymore. We miss having an adversary so much we started shit with Iran, who is dabbing on America on Twitter, which is embarrassing for everyone. But this distillation thing. And so, Steve, I assume what you mean is like the models, like Chinese models training off of outputs from ChatGPT and Claude, right?
That's what they claim, yeah.
Yeah. So here's two things with this. One, you're completely right. Who gives a fuck? Oh, suddenly the stealing machine doesn't like when it's stolen from? What are you going to do about it, bitch? Why don't you distill from other people? It isn't theft. It's ridiculous. And it's so silly as well because their argument is, well, it's our models that created the outputs and thus we own them. It's like, okay, well, who created the shit you trained on?